GeneForge: AI Driven Genetic Circuit Design

Overview

GeneForge is a project aimed at developing a robust framework for generating and modeling genetic circuits. The ultimate goal is to create a model capable of designing and editing genetic circuits given an initial and target cell states (e.g. from RNA-seq data).

Project Structure

├── src/
│   ├── circuits/         # for generating circuits
│   ├── sboll_llm/        # sub-project for generating sbol compliant circuits designs from natural language
|       ├── data/         # for downloading and processing datasets
│       ├── repositories/ # for interacting with parts repos
│       ├── train/        # training scripts
├── notebooks/
├── docs/             # references and background material
├── .gitignore
├── README.md
├── requirements.txt

Goals

Develop a Generative Model for Genetic Circuit Design:

• Train models to learn and generate valid genetic circuits using parts repositories such as SynBioHub and iGEM and language modeling techniques.
• Validate the ability to model and generate valid circuits.

Generate Circuits Based on Initial and Target Cell States:

• Use expression data (e.g., RNA-seq) to design genetic circuits that transition a cell from an initial expression state to a target expression state.
• Generate a large quantity of pseudo-random circuit design.
• Combine circuit design simulations (e.g. libSBML) with perturbation simulations (e.g., GEARs, GeneFormer).

References and Background Material

A bibliography of related publications can be found in the docs folder.

Future Work

• Learning/tuning scheme for component parameter optimization (e.g. binding constants, degradation and production rates)
• Graph representations and graph kernels
• Circuit simulations -> perturbation-seq simulations